Research On Performance Monitoring And PID Tuning Method For PID Control System Based On Evidential K-NN Classifier

PID controller is widely used in thermotechnical control of modern thermal power generation.In order to improve the safety and economy of thermal power generation process and realize the automation and intelligence of thermotechnical control,improving the performance of PID control system is the key way.In this paper,the performance monitoring and PID tuning method for PID control system based on evidential K-NN classifier is studied,which can realize performance monitoring,closed-loop identification and intelligent tuning of PID controller parameters under imprecise and uncertain data conditions.The main contents and achievements of this paper are as follows:(1)A performance monitoring method for PID control system based on pattern migration trend is proposed.The state and performance of the control system are characterized by pattern.Under the framework of evidence theory,multiple deterministic performance indicators such as rise time are fused into a mass function indicator in evidence representation which is used as the representation of the system pattern,and the size and change trend of the mass function are utilized as the criteria of the system pattern and the pattern migration trend.Using evidential K-NN classifier,under the condition of imprecise and uncertain data,k nearest neighbor training set samples are selected to construct and fuse evidence,and the above mass function is calculated.Simulation results show that this method can not only identify the current pattern of the system,but also calculate the migration trend of the system pattern,which solves the deficiency that the performance monitoring method based on indicators can only obtain the current system state but cannot obtain the state evolution trend.(2)An adaptive k-value based close-loop identification method of neighbor sample fusion based on performance indicators is proposed.This method is based on the off-line training set samples,and the transfer functions corresponding to the k nearest neighbor samples of the current system to be identified are used to perform the addition operation of the transfer functions.Four fusion methods are discussed in this chapter which are:(a)Equally weighted system closed-loop transfer function fusion method.(b)Distance distribution weighted system closed-loop transfer function fusion method.(c)Equally weighted system open-loop transfer function fusion method.(d)Distance distribution weighted system open-loop transfer function fusion method.In the above four fusion methods,the principle is to minimize the error of the closed-loop step response data of the system corresponding to the identification result and the current system,and the optimal nearest neighbor amount k is adaptively selected,and the optimal solution is taken as the closed-loop identification result.Simulation results show that this method does not need online optimization identification of the control system,which greatly shortens the on-line calculation time.(3)A neighborhood equivalent PID “pull-back” tuning method based on system performance indicator is proposed.In this method,based on the off-line training set generated by perturbing system PID parameters and process delay time,by locking the nearest neighbor sample information of the current system,the change of the system performance indicators caused by the change of the process of the current system is equivalent to the change of the equivalent performance indicators caused by the change of the PID parameter of the nearest neighbor sample system.According to the difference of PID parameters between the expected system and the nearest neighbor sample system,the current system controller parameters are adjusted reversely to pull the current system performance back to the expected system performance.The results show that this method does not need an accurate process model,and it can effectively realize PID controller tuning and greatly simplify the on-line tuning process.